NTU RGB + D 120 动作识别数据集 描绘骨架(python)
“ NTU RGB + D”和“ NTU RGB + D 120”。
“ NTU RGB + D”包含60个动作类别和56,880个视频样本。
“ NTU RGB + D 120”通过添加另外60类和另外57,600个视频样本来扩展“ NTU RGB + D”，即“ NTU RGB + D 120”总共有120类和114,480个样本。
这两个数据集都包含每个样本的RGB视频，深度图序列，3D骨骼数据和红外（IR）视频。每个数据集由三台Kinect V2摄像机同时捕获。
RGB视频的分辨率为1920x1080，深度图和IR视频的分辨率均为512x424，并且3D骨骼数据包含每帧25个人体关节的3D坐标。
有那些动作？
数据集的意思
参考这篇文章：https://blog.csdn.net/hsclyy/article/details/102809083
代码：
# -*- coding:utf-8 -*-
import os
import numpy as np
from mpl_toolkits.mplot3d import Axes3D
import matplotlib.pyplot as plt
trunk_joints = [0, 1, 20, 2, 3]
arm_joints = [23, 24, 11, 10, 9, 8, 20, 4, 5, 6, 7, 22, 21]
leg_joints = [19, 18, 17, 16, 0, 12, 13, 14, 15]
body = [trunk_joints, arm_joints, leg_joints]
# Show 3D Skeleton with Axes3D for NTU RGB+D
class Draw3DSkeleton(object):
def __init__(self, file, save_path=None, init_horizon=-45,
init_vertical=20, x_rotation=None,
y_rotation=None, pause_step=0.2):
self.file = file
self.save_path = save_path
# if not os.path.exists(self.save_path):
# os.mkdir(self.save_path)
self.xyz = self.read_xyz(self.file)
self.init_horizon = init_horizon
self.init_vertical = init_vertical
self.x_rotation = x_rotation
self.y_rotation = y_rotation
self._pause_step = pause_step
def _read_skeleton(self, file):
with open(file, 'r') as f:
skeleton_sequence = {}
skeleton_sequence['numFrame'] = int(f.readline())
skeleton_sequence['frameInfo'] = []
for t in range(skeleton_sequence['numFrame']):
frame_info = {}
frame_info['numBody'] = int(f.readline())
frame_info['bodyInfo'] = []
for m in range(frame_info['numBody']):
body_info = {}
body_info_key = [
'bodyID', 'clipedEdges', 'handLeftConfidence',
'handLeftState', 'handRightConfidence', 'handRightState',
'isResticted', 'leanX', 'leanY', 'trackingState'
]
body_info = {
k: float(v)
for k, v in zip(body_info_key, f.readline().split())
}
body_info['numJoint'] = int(f.readline())
body_info['jointInfo'] = []
for v in range(body_info['numJoint']):
joint_info_key = [
'x', 'y', 'z', 'depthX', 'depthY', 'colorX', 'colorY',
'orientationW', 'orientationX', 'orientationY',
'orientationZ', 'trackingState'
]
joint_info = {
k: float(v)
for k, v in zip(joint_info_key, f.readline().split())
}
body_info['jointInfo'].append(joint_info)
frame_info['bodyInfo'].append(body_info)
skeleton_sequence['frameInfo'].append(frame_info)
return skeleton_sequence
def read_xyz(self, file, max_body=2, num_joint=25):
seq_info = self._read_skeleton(file)
data = np.zeros((3, seq_info['numFrame'], num_joint, max_body)) # (3,frame_nums,25 2)
for n, f in enumerate(seq_info['frameInfo']):
for m, b in enumerate(f['bodyInfo']):
for j, v in enumerate(b['jointInfo']):
if m < max_body and j < num_joint:
data[:, n, j, m] = [v['x'], v['y'], v['z']]
else:
pass
return data
def _normal_skeleton(self, data):
# use as center joint
center_joint = data[0, :, 0, :]
center_jointx = np.mean(center_joint[:, 0])
center_jointy = np.mean(center_joint[:, 1])
center_jointz = np.mean(center_joint[:, 2])
center = np.array([center_jointx, center_jointy, center_jointz])
data = data - center
return data
def _rotation(self, data, alpha=0, beta=0):
# rotate the skeleton around x-y axis
r_alpha = alpha * np.pi / 180
r_beta = beta * np.pi / 180
rx = np.array([[1, 0, 0],
[0, np.cos(r_alpha), -1 * np.sin(r_alpha)],
[0, np.sin(r_alpha), np.cos(r_alpha)]]
)
ry = np.array([
[np.cos(r_beta), 0, np.sin(r_beta)],
[0, 1, 0],
[-1 * np.sin(r_beta), 0, np.cos(r_beta)],
])
r = ry.dot(rx)
data = data.dot(r)
return data
def visual_skeleton(self):
fig = plt.figure()
ax = Axes3D(fig)
ax.view_init(self.init_vertical, self.init_horizon)
plt.ion()
data = np.transpose(self.xyz, (3, 1, 2, 0)) # 调换索引值，将x下标的索引值换到最后，将max_body索引值换到前面来
# data rotation
if (self.x_rotation is not None) or (self.y_rotation is not None):
if self.x_rotation > 180 or self.y_rotation > 180:
raise Exception("rotation angle should be less than 180.")
else:
data = self._rotation(data, self.x_rotation, self.y_rotation)
# data normalization
data = self._normal_skeleton(data)
# show every frame 3d skeleton
for frame_idx in range(data.shape[1]):
plt.cla()
plt.title("Frame: {}".format(frame_idx))
ax.set_xlim3d([-1, 1])
ax.set_ylim3d([-1, 1])
ax.set_zlim3d([-0.8, 0.8])
x = data[0, frame_idx, :, 0]
y = data[0, frame_idx, :, 1]
z = data[0, frame_idx, :, 2]
for part in body:
x_plot = x[part]
y_plot = y[part]
z_plot = z[part]
ax.plot(x_plot, z_plot, y_plot, color='b', marker='o', markerfacecolor='r')
ax.set_xlabel('X')
ax.set_ylabel('Z')
ax.set_zlabel('Y')
if self.save_path is not None:
save_pth = os.path.join(self.save_path, '{}.png'.format(frame_idx))
plt.savefig(save_pth)
print("The {} frame 3d skeleton......".format(frame_idx))
ax.set_facecolor('none')
plt.pause(self._pause_step)
plt.ioff()
ax.axis('off')
plt.show()
if __name__ == '__main__':
# test sample
sk = Draw3DSkeleton("D:\BaiduNetdiskDownload\Raw_Skeleton_S01-S17\skeleton+D0-30000\S001C001P001R001A050.skeleton")
sk.visual_skeleton()